Delegated Authority Without Accountability: The Governance Discontinuity Created by Agentic AI

Artificial intelligence is increasingly moving beyond prediction, recommendation, and content generation into the initiation and execution of consequential action. Systems capable of selecting tasks, sequencing workflows, using external tools, adapting to changing conditions, and acting without continuous human direction occupy a different functional position from conventional software tools. This paper argues that such systems should be treated as non-human operational actors for governance purposes, without implying consciousness, legal personhood, moral agency, or independent authority. The paper distinguishes technical capability, system permission, and institutional authority, showing how capability combined with access can produce effective operational power without an equivalent act of legitimate delegation. It defines the resulting governance discontinuity as the structural gap between autonomous operational capability and the authority, constraint, interruption, provenance, and accountability arrangements required to govern its exercise. From this discontinuity arise recurring structural failure modes, including authority drift, silent execution, escalation failure, responsibility diffusion, contextual misalignment, compositional overreach, cumulative drift, coordination failure, provenance collapse, and decay of meaningful human oversight. The paper argues that incremental additions to model governance, access control, monitoring, policy, lifecycle assurance, and human-in-the-loop review remain necessary but may be insufficient because they preserve governance assumptions centred on tools and human-initiated action. It identifies minimum requirements for governable agency: declared purpose, explicit delegation, bounded and context-sensitive autonomy, risk-sensitive operation, mandatory interruption and escalation, governance provenance, cumulative and compositional awareness, governance across coordination pathways, institutional control of governance constraints, and accountable human anchoring. The paper concludes that as consequential action becomes increasingly executed by non-human operational actors, governance must remain continuously coupled to both the delegation of authority and the execution of action.